benchgap
Calibration

MMLU → AA-Omniscience Accuracy

AA-Omniscience Accuracy is estimated from MMLU with a inverse Michaelis–Menten curve fitted on 5 models measured on both: y = 0.03336·(x − 0.0000) / (0.0000 + 1.0081 − x), R² = 0.98, cross-validated error 1.8 pp. It is used for 2 estimates.

Estimated modelMMLUAA-Omniscience AccuracySource
LongCat-Flash-Lite-Sparse85.3%18.4%estimated ± 1.8 pp, medium confidence
o3-mini86.9%20.8%estimated ± 1.8 pp, medium confidence